How Close Are We to Autonomous Driving?
Posted by Troy Larson on Thursday, October 30, 2014 at 12:00 AM
By Troy Larson / October 30, 2014
Comment
I have always been intrigued by Google’s self-driving car -- this meant there was a possibility of a blind person like myself being able to travel by their own car to wherever they wanted to go. Or someone who was fully sighted could be doing other activities such as browsing the internet while the car takes them to their destination.
At the same time, I was also skeptical of the idea -- questions like: “What if the car fails to stop and runs into a person while they were crossing the street?” In that case, it meant that a blind person or someone who wasn’t focused on the road wouldn’t be able to safely drive the car if such an occurrence happened.
A recent article confirmed some of my doubts. These self-driving cars would need a computer far more intelligent than what is currently available. It may take several years before this can be accomplished if ever.
The biggest hurdle for these cars is they can’t function unless the road is mapped. This would mean mapping all four million miles of the public roads here in the U.S., plus all driveways and off-road trails that anyone would want to take. You would think that with all that Google has accomplished with their mapping technologies that they would be able to do this efficiently. Unfortunately, it doesn’t seem to be all that simple.
When Google maps a road for their self-driving car, they first have a vehicle fitted with a bank of sensors drive repeatedly over the roadway to be mapped, recording all fixed objects such as streetlights and stop signs along with lane markings and crosswalks. After which, the data is downloaded and analyzed extensively by humans and computers to ensure that all real life objects have been mapped. After which, the map is then loaded into the car’s memory before a journey. With the map in the car’s memory, the car which uses a generic PC running Ubuntu Linux, can spend more of its resources to tracking moving objects, like other cars.
Currently, Google has a few thousand miles of roads mapped for their car, the majority of which are around the corporate headquarters in Mountain View, California. Google has claimed repeatedly that their car has driven safely over 700 thousand miles, but it is over the same few thousand miles of mapped roads.
Once Google had its massive database of maps in place. The challenge would be to keep them current. A scenario where they would need to be updated is if a traffic light was to be installed or there was significant construction. Usually, there’s sufficient time to build a traffic light. However, sometimes a construction company may put up a portable traffic light. In that case, there isn’t sufficient time to update the database.
Another issue with the car is it can’t tell the difference between a rock and a piece of crumpled up newspaper in the road and will try to go around both of them. Not only that, it can’t tell what color the traffic light is if the camera is blinded by the sun.
For now, it looks like Google’s self-driving car may not be an actual reality anytime soon. It seems like that we may have reached the limits of technology since the car can’t perform some of the functions that you would normally do while driving. The idea of a self-driving car is still that of science fiction. All we can do is keep dreaming.
Photos via: Google -- Volvo
Post a Comment